In October 2026, several open-weight AI models were released, highlighting the growing trend of making model parameters publicly available. Meta's Muse Spark Open, a massive 816B parameter model, requires approximately 1 TB of RAM to run locally, demonstrating the significant hardware demands of these models. Other releases include DeepSeek's V4.1 Flash 1017 with 552B backbone parameters, IBM's Granite Switch 4.1 8B, and Qwen3.8-27B, offering a range of sizes and capabilities for researchers and developers. AI
IMPACT Provides insights into the hardware requirements and licensing of open-weight models, aiding developers in selecting and deploying AI.
RANK_REASON Article details multiple new open-weight model releases and discusses the technical aspects of running them, fitting the research category. [lever_c_demoted from research: ic=1 ai=1.0]
- Apple Inc.
- DeepSeek
- Engadget
- Hugging Face
- IBM
- Meta
- Mlx
- modelscope
- Muse Spark Open
- NVFP4
- Open Source Initiative
- Qwen3.8-27B
- Toolbit
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →